Šimon Hudínek

CVS emission benches are humidity-sensitive: condensation in the dilution tunnel invalidates a test run. I owned the loop end to end — framed the problem, designed the measurement campaign, instrumented the test cell, built the thermodynamic model, validated it against PEMS data, and shipped the operator GUI that flags condensation-prone runs before they start. The panel is that pre-check tool rebuilt in the browser.

Too little AdBlue and NOx escapes; too much and ammonia slips through — which is why production exhaust systems carry an extra ammonia-slip catalyst (ASC) behind the SCR. My thesis NMPC previews load steps 8 minutes ahead and holds NH3 slip below 1 ppm, so the ASC can be deleted — traded against some NOx conversion at partial load. With no hardware to fall back on, every claim had to survive validation against the XMR monolith reactor simulator. The panel shows the system and the benchmark.

As CTO of an 8-engineer semester team I owned the architecture, split the hardware and software branches, and integrated them into one system delivered on schedule. The stack: first-principles model, nonlinear MPC with EKF state estimation, NI-DAQmx acquisition, and a PySide6 operator app — stable closed-loop control on the real rig. The panel is that operator app, replaying a closed-loop run at 6× real time.

Off the clock I build the things I use: an ESP32 circadian lighting cabinet driven by a local server (no cloud, running as a daily-driver), a capacitive-sensing smart bottle that logs hydration patterns, and a laboratory test rig brought back from a hardware failure by remapping its wiring and rebuilding the LabVIEW control logic. Click a photo below for the full story of each.